2x RTX 3080 20GB
4 records across 1 models (quants: NVFP4 / f16 KV, NVFP4 / q8_0 KV).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2x RTX 3080 20GB | qwen3.5-arch 27B | NVFP4 / f16 KV | llama.cpp (CUDA) | 50.6 | – | – | – | 30000 | 2026-10-04 | community | – | source |
| 2x RTX 3080 20GB | qwen3.5-arch 27B | NVFP4 / q8_0 KV | llama.cpp (CUDA) | 48.7 | – | – | – | 30000 | 2026-10-04 | community | – | source |
| 2x RTX 3080 20GB | qwen3.5-arch 27B | NVFP4 / f16 KV | llama.cpp (CUDA) | 44.0 | – | – | – | 90000 | 2026-10-04 | community | – | source |
| 2x RTX 3080 20GB | qwen3.5-arch 27B | NVFP4 / q8_0 KV | llama.cpp (CUDA) | 39.4 | – | – | – | 90000 | 2026-10-04 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085